Data-Driven Lead Optimisation for Drug Discovery

研究问题：这项研究提供了一个领先的优化工具，允许化学家探索以前为特定目标工作的化学空间。然后，这个工具应该能够建议应该用推理探索的分子或化学空间区域。这应该允许化学家有一个更好的理解，为什么一个特定的分子被选择探索一个特定的目标。将用于该工具的主要分子表示是简化图，这是一种已简化到关键交互节点的图形表示。可视化工具应该是一个交互界面，化学家可以有一个化学空间的整体视图，同时也允许他们了解每个分子的复杂细节。原始方法：迄今为止所采用的方法是将化学结构简化为简化图。这些简化图已经制作完成，因此用户可以自定义它们，因为它们可以设置不同的参数和不同的定义，这取决于他们正在寻找的内容和他们找到的键。下一步是使用python模块RDKit找到最大公共子结构（MCS），包括连接和断开连接的版本。这些MCS被发现，因此一些聚类技术可以发生，因为它们是基于使用谷本系数方程中的MCS发现的相似性或不相似性分数。然后执行几种不同的聚类技术，并应用聚类有效性技术，以便为该数据集建立最佳聚类。发现这些簇的核心简化图有助于可视化技术。然后产生一个可视化，总结了这些信息，已经工作的化学空间。从这里需要建立一种方法，使一个简化的图转换回化学图。这个化学图将在一个活性模型中运行，看看这个分子应该有多好，如果新的活性预测是充分的，那么这个分子将被作为一个建议提出。然后，可视化将被用来支持这个建议。

Research questions:This research provides a lead optimisation tool that allows chemists to explore the chemical space that has previously been worked in for a certain target. This tool should then be able to suggest molecules or areas of chemical space that should be explored with reasoning. This should allow the chemist to have a greater understanding as to why a certain molecule has been selected to explore for a certain target. The main molecular representation that will be used for the tool is reduced graphs, which is a graphical representation that has been reduced down to the key interacting nodes. The visualisation tool should be an interactive interface that the chemists can have an overall view of the chemical space whilst also still allowing them to complex details of each molecule.Original methodology:The methodology that has been done so far has been to reduce the chemical structures down into reduced graphs. These reduced graphs have been made so they are customisable for the user as they can set different parameters and different definitions depending on what they are looking for and what they find key. The next step was to then find the maximum common substructure (MCS), both the connected and disconnected versions, using a python module RDKit. These MCS' are found so that some clustering techniques can occur as they are based upon similarity or dissimilarity scores which are found using the MCS in the Tanimoto coefficient equation. Several different clustering techniques are then performed and cluster validity techniques are applied in order to establish the best clusters for that dataset. A core reduced graph of these clusters are found to aid the visualisation technique. A visualisation is then produced that summaries this information, the chemical space that has been worked in. From here a method needs to be established that enables a reduced graph to be converted back into a chemical graph. This chemical graph will be ran in an activity model to see how well this molecule should perform and if the new prediction of activity is adequate then the molecule will be put forward as a suggestion. The visualisation will then be used to back up this suggestion.